(1) THOUSANDS SIGN AUTHORS GUILD LETTER CALLING ON AI INDUSTRY TO PROTECT WRITERS. Over 9000 authors, including genre-recognizable names like Margaret Atwood, Michael Chabon, Carmen Machado, Joe Hill, Edward M. Lerner, Brendan DuBois, Terri Windling, Matthew Kressel, Sean Wallace, and Cecilia Tan, have signed an open letter from the Authors Guild to the CEOs of companies developing AI generative software to not use their members’ work without consent, credit, and compensation when developing their systems.
The Authors Guild, the leading professional organization for writers in the United States, has submitted an open letter to the CEOs of prominent AI companies, including OpenAI, Alphabet, Meta, Stability AI, IBM, and Microsoft. The letter calls attention to the inherent injustice of building lucrative generative AI technologies using copyrighted works and asks AI developers to obtain consent from, credit, and fairly compensate authors.
More than 9,000 writers and their supporters have signed the letter including luminaries such as Dan Brown, James Patterson, Jennifer Egan, David Baldacci, Michael Chabon, Nora Roberts, Jesmyn Ward, Jodi Picoult, Ron Chernow, Michael Pollan, Suzanne Collins, Margaret Atwood, Jonathan Franzen, Roxane Gay, Celeste Ng, Louise Erdrich, Viet Thanh Nguyen, George Saunders, Min Jin Lee, Andrew Solomon, Rebecca Makkai, Tobias Wolff, and many others.
The open letter emphasizes that generative AI technologies heavily rely on authors’ language, stories, style, and ideas. Millions of copyrighted books, articles, essays, and poetry serve as the foundation for AI systems, yet authors have not received any compensation for their contributions. These works are part of the fabric of the language models that power ChatGPT, Bard, and other generative AI systems. Where AI companies like to say that their machines simply “read” the texts that they are trained on, this is inaccurate anthropomorphizing. Rather, they copy the texts into the software itself, and then they reproduce them again and again.
Maya Shanbhag Lang, president of the Authors Guild, said, “The output of AI will always be derivative in nature. AI regurgitates what it takes in, which is the work of human writers. It’s only fair that authors be compensated for having ‘fed’ AI and continuing to inform its evolution. Our work cannot be used without consent, credit, and compensation. All three are a must.”…
We, the undersigned, call your attention to the inherent injustice in exploiting our works as part of your AI systems without our consent, credit, or compensation.
Generative AI technologies built on large language models owe their existence to our writings. These technologies mimic and regurgitate our language, stories, style, and ideas. Millions of copyrighted books, articles, essays, and poetry provide the “food” for AI systems, endless meals for which there has been no bill. You’re spending billions of dollars to develop AI technology. It is only fair that you compensate us for using our writings, without which AI would be banal and extremely limited.
We understand that many of the books used to develop AI systems originated from notorious piracy websites. Not only does the recent Supreme Court decision in Warhol v. Goldsmith make clear that the high commerciality of your use argues against fair use, but no court would excuse copying illegally sourced works as fair use….
Last week, SFWA signed the “Creators Call for Action on Artificial Intelligence (AI) Copyright Exceptions,” a joint action from 24 creator-led organizations delivered to the European Union and United States governing bodies. This letter addresses the harm already caused to creators by AI companies’ manipulations of exceptions to copyright enacted by the 2019 European Union Directive on Copyright in the Digital Single Market. The details of this issue are well explained on the National Writers Union (NWU) website.
We felt it was important to sign onto this particular joint call for a number of reasons:
Working with other organizations to monitor and respond to developments both in generative AI and in the legal environment around it is essential to advocating for concrete protections and restitution for writers.
Responding to the unique issues that AI presents to creative copyrights is a global effort, not one limited to Silicon Valley. It’s important to ensure that writers’ rights are respected around the world, so corporations who may try to relocate their operations rather than compensate creators fairly will have no place to go.
This letter, written by the NWU, is one such global effort, and we encourage everyone to read the letter in full to better understand some of the important copyright issues regarding the use of AI/ML applications. You are welcome to distribute the joint call and spread the word. Access the PDF of the complete joint call for action here.
A South Carolina school board meeting, in which community members railed against an African American culture writer’s award-winning memoir about racial injustice, featured a special guest appearance: Ta-Nehisi Coates, the famed author in question.
On Monday evening, the Lexington-Richland District 5 School Board met to discuss the outrage concerning Coates’ 2015 nonfiction bestseller, Between the World and Me, which has repeatedly caused political literary mayhem among reactionary right-wing communities and been placed on book ban lists.
In February, after getting approval from higher-ups, an AP Language teacher at Chapin High School conducted a lesson involving Between the World and Me. The book, written as an essay to Coates’ son to prepare him for the life he will live as a Black man, details personal accounts of Coates’ life and his first-hand experiences with racism. However, the lesson was shut down and the book was removed from the course after students filed a complaint claiming the book made them feel “guilty for being white,” local news outlet CBS 19 Columbia reported.
According to footage obtained by CBS 19, a slew of people wearing blue rallied in support for the book and for academic freedom during the board hearing. And Coates sat in the back of the room next to the teacher who assigned the book as a sign of solidarity….
… The board did not conduct a vote after public discussion.
In a statement to The Daily Beast, Lexington-Richmond District 5 wrote that it is “important to understand” that Between the World and Me “is not banned in our school district.”…
(4) INNOCENCE ASSERTED. Michele Lundgren, wife of sff artist Carl Lundgren, charged by Michigan authorities this week with numerous felonies as a fake Trump elector, gave an interview to a reporter in which she says she was duped. “I was an innocent little bystander in the whole thing thinking I was doing my civic duty.”
However, if you watch this Detroit Free Press video from December 15, 2020 — https://www.youtube.com/watch?v=P_NgLQxMV9c&t=206s — it shows her just after the Michigan fake electors were turned away from the State House, standing next to Ian Northon, an attorney, nodding vigorously as he explains his theory of why the GOP fake electors should have been let in. (Thanks to Kathryn Cramer for the link.)
Michele Lundgren, left, in news video covering 2020 attempt to deliver fake Trump electors.
… As with all of the stories in the Liaden Universe® series, Salvage Right is a sci-fi space opera story. But are there any other genres at work in this story as well?
Sharon: Salvage Right is as pure a space opera as we’ve written in a while. It was fun to let all the stops out.
Steve: We also draw from regency romance and comedy of manners fiction; almost every time a character bows that’s tip of the hat to Georgette Heyer!
Are there any writers, or stories, who had a big influence on Salvage Right but not on anything else you’ve written?
Steve: I don’t think so; nothing recent, certainly, and with Salvage Right being a merge of story lines, it would be hard to filter one new factor in, I think.
Sharon: Harlan Ellison’s short story “I Have No Mouth And I Must Scream”; The Prestige, the novel by Christopher Priest….
(6) O CAPTAIN, MY CAPTAIN. Kimberly Unger is at San Diego Comic-Con. From her hotel room window she was able to take a great photo of this bit of Star Trek publicity.
(7) TRACING OPPENHEIMER’S FOOTSTEPS IN SOUTHERN CALIFORNIA. William Deverell takes readers on a historic tour of a Caltech neighborhood significant to the history of J. Robert Oppenheimer in “The Pasadena Project” at Alta Online.
It’s only a couple of hundred yards from our house to the heart of the Caltech campus, less than a 10-minute walk, an environment our dogs never tire of exploring. Devoted mostly to classrooms, dorms (“houses,” in Caltech parlance), and research labs, the campus is almost always quiet, and if you know where to look, there are places on campus that are little changed from when Oppenheimer and members of his eventual Los Alamos team worked there.
Hiding in plain sight, innocuous and out of place, is a small Spanish revival home, smack-dab in the middle of everything else. This is the Tolman-Bacher House. Built around the same time as our home, which it resembles, the Tolman-Bacher House was the residence of Richard Tolman and his wife, Ruth (Louise Lombard plays her in the film). Richard Tolman had come to Caltech in 1922 as a professor of physics. He was a giant at Caltech and, later, in the Manhattan Project. Oppenheimer (played in the film by Irish actor Cillian Murphy) often stayed with the Tolmans when he’d come to Caltech from Berkeley during many a spring term in the 1930s. His brother, Frank, was a graduate student at Caltech then….
… Being a Nolan screenplay, the story is told in nonlinear fashion. Cillian Murphy, with his open, seemingly guileless expression, is completely convincing as the scientist known as the father of the atomic bomb who, after building it, counseled against its use and made many enemies in the process. But no one can get inside the head of a genius—be it a painter or a composer or a brilliant scientist, so we don’t leave the theater with a feeling of knowing what Oppy was all about, except on the surface. (There is even a glimpse of Albert Einstein, played by that wonderful actor Tom Conti.)…
(9) FRANK WALLER (1957-2023.) Longtime LASFS member Frank Waller died July 18 at the age of 66. He had quadruple bypass surgery on his heart in May, but had gone through rehab and was out of the hospital. Frank joined the club in 1988. He is survived by his sister, Beth and his brother, Joe.
(10) MEMORY LANE.
2005 – [Written by Cat Eldridge from a choice by Mike Glyer.]
Graham Joyce wrote our Beginning this time. Selecting his best novel is a futile exercise as everything is fantastically good but I’ll single out Some Kind of Fairy Tale and The Tooth Fairy as the ones I found the most interesting reads.
No Hugos, not even a nomination, but he’s won a few BFAs and even one WFA for The Facts of Life novel.
Mike picked The Limits of Enchantment which was published eighteen years ago by Artia Books in the States with the cover illustration by David Sacks. It was nominated for a World Fantasy Award.
And now for our Beginning…
Prologue
If I could tell you this in a single sitting then you might believe all of it, even the strangest part. Even the part about what I found in the hedgerow. If I could unwind this story in a single spool, or peel it like an apple the way Mammy would with her penknife in one unbroken coil, juice a-glistening on the blade, then you might bite in without objection.
But Mammy always said we have lost the art of Listening. She said we live in an age when everyone chatters and no one takes heed, and that, she said, is not a good time in which to live.
And while I offer you my story unbroken, like the apple peel, it hangs by a fibre at every turn of the knife. When you come to know the nature of the teller of this tale you may have good reason to doubt both. You may suspect the balance of my mind and you may condemn my position. You may start to disbelieve.
Perhaps I once was mad. Briefly. Perhaps that much is true. And this, in an age where we no longer have the patience to listen, may cause you to break off, to give up on me, to turn away. A young woman has so little of interest to offer, after all. A young woman of unsteady temper, even less.
What they did to Mammy they tried to do to me. They released the dogs. And when it comes to telling how it was done, I only ask this: when doubt wrinkles your brow; when incomprehension clouds your eyes; when distaste rests like a rank fog on your lips, then think how we few have held our tongues for so long. How we have choked back the truth. How we have burned in our hearts rather than risk the telling. And when you feel most far from me, then at that moment listen hard. Not to your thoughts, which will mislead you, nor to your heart, which will lie, but to the voice behind the voice, and trust the tale and not the teller.
(11) TODAY’S BIRTHDAYS.
[Compiled by Cat Eldridge.]
Born July 20, 1924 — Lola Albright. Though she’s best remembered best known for playing the sultry singer Edie Hart, the girlfriend of private eye Peter Gunn, she did do some genre performances. She’s Cathy Barrett, one of the leads in the Fifties film The Monolith Monsters. Television was really her home in the Fifties and Sixties. She was on Tales of Tomorrow as Carol Williams in the “The Miraculous Serum” episode, Nancy Metcalfe on Rocket Squad in “The System” episode, repeated appearances on the various Alfred Hitchcock series, and even on The Man from U.N.C.L.E. in the episodes released as the feature length film The Helicopter Spies. She was Azalea. (Died 2017.)
Born July 20, 1930 — Sally Ann Howes. She is best known for the role of Truly Scrumptious in Chitty Chitty Bang Bang. She was in Brigadoon as Fiona McLarenat New York City Center Light Opera Company, and in Camelot as Guenevere at St. Louis Municipal Opera. She was even in The Hound of the Baskervilles as Laura Frankland which has a certain Starship Captain as George Stapleton. (Died 2021.)
Born July 20, 1931 — Donald Moffitt. Author of the Baroness thriller series, somewhat akin to Bond and Blaise, but not quite. Great popcorn literature. Some SF, two in his Mechanical Sky series, Crescent in the Sky and A Gathering of Stars, another two in his Genesis Quest series, Genesis Quest and Second Genesis, plus several one-offs. (Died 2014.)
Born July 20, 1938 — Diana Rigg, née Dame Enid Diana Elizabeth Rigg. Emma Peel of course in The Avengers beside Patrick Macnee as a John Steed. Best pairing ever. Played Sonya Winter in The Assassination Bureau followed by being Contessa Teresa “Tracy” Draco di Vicenzo Bond on On Her Majesty’s Secret Service. By the Eighties, she’s doing lighter fare such as being Lady Holiday in The Great Muppet Caper and Miss Hardbroom in The Worst Witch, not to mention The Evil Queen, Snow White’s evil stepmother in Snow White. Now she would get a meaty role in Game of Thrones when she was Olenna Tyrell. Oh and she showed up in Dr. Who during the Era of the Eleventh Doctor as Mrs. Winifred Gillyflower in the “The Crimson Horror” episode. (Died 2020.)
Born July 20, 1947 — Michael “Mike” Gilbert. A fan artist in the late ’60s in Locus and other fanzines as well as an author, and publishing professional who won a Hugo Award for Best Fan Artist at the first Noreascon. His wife Sheila was the co-publisher of DAW Books, and Mike worked in both editorial and art capacities at DAW, and was one of their primary first readers. He died of complications following open-heart surgery. (Died 2000.)
Born July 20, 1949 — Guy H. Lillian III, 74. Letterhack and fanzine publisher notable for having been twice nominated for a Hugo Award as best fan writer and rather amazingly having been nominated twelve straight times for the Hugo for best fanzine for his Challenger zine, unfortunately never winning. As a well-fan of Green Lantern, Lillian’s name was tuckerized for the title’s 1968 debut character Guy Gardner.
Born July 20, 1959 — Martha Soukup, 64. The 1994 short film Override, directed by Danny Glover, was based on her short story “Over the Long Haul”. It was his directorial debut. She has two collections, Collections Rosemary’s Brain: And Other Tales of Wonder and The Arbitrary Placement of Walls, both published in the Nineties. She won a Nebula Award for Best Short Story for “A Defense of the Social Contracts”. “The Story So Far” by her is available as the download sample at the usual suspects in Schimel’s Things Invisible to See anthology if you’d liked to see how she is as a writer.
(13) OCTOTHORPE. In episode 88 of the Octothorpe podcast, “Someone Somewhere Must Know What’s Going On”, John Coxon is on brand, Alison Scott wants to do Mark, and Liz Batty had a few twinges.
We re-introduce ourselves for anyone who hasn’t listened before, and then we dive into in-jokes, regular segments, waffling, and all the other things that have firmly cemented us as “a podcast that people can listen to”. (COVID, Eastercon, Hugo Awards, Chengdu, Clarke Awards, Arkham Horror, and—checks notes—cycling.)
(14) VOYAGE TO ARCTURUS REFERENCED ON BANDCAMP. [Item by Steve French.] Back in 1974, British experimental folk musician Mike Cooper released his ‘landmark’ album, Life and Death in Paradise. The opening track was apparently influenced by David Lindsay’s classic 1920 SF novel A Voyage to Arcturus, which influenced Lewis, Tolkien and more recently, Pullman. “A Reissue of ‘Life and Death in Paradise’ Brings Mike Cooper’s Music to a New Generation” at Bandcamp Daily.
The opener, “Rocket Summer,” like the songs on Trout Steel (inspired by Richard Brautigan’sTrout Fishing in America) was born of Cooper’s “esoteric” choice in literature—in this case, a sci-fi book called A Voyage to Arcturus. Aside from its narrative of interstellar travel, it’s filled with “lots of references to sound and color” which were attractive to Cooper.
The album has just been re-issued, prompting Bandcamp to give Cooper the full-on ‘feature’ treatment.
…Around 2010, Rice wanted to make the runs harder, and he thought wearing a mask and black clothing would do the trick. When he remembered that parts of the Star Wars franchise were filmed in Death Valley, he got the idea to dress up as the villain of the series.
Rice, who edits a cryptocurrency trade publication, has done the Darth Valley run most years since then, with breaks during the coronavirus pandemic and a cross-country move. Sometimes other runners join him – occasionally in a Chewbacca costume….
In a Halloween-store Darth Vader costume, complete with a helmet, Jon Rice tries to make his near-annual “Darth Valley” run as difficult as possible.
On Sunday, the New Mexico resident completed his run as temperatures neared 130 degrees Fahrenheit. https://t.co/bdNlQz8CNE
— The Washington Post (@washingtonpost) July 18, 2023
The “Wilhelm scream” is arguably the most recognizable stock sound effect in the history of film and television, having been used in everything from mega-franchises like “Star Wars,” “Indiana Jones” and “Die Hard,’ to beloved TV shows from “X-Files” to “SpongeBob SquarePants” to “Game of Thrones” and beyond.
If the mere mention of its name doesn’t immediately make the sound play in your head, you may recognize it from this scene in the movie that made it popular, 1977’s “Star Wars: Episode IV — A New Hope,” when the effect is used for a stormtrooper that falls off a ledge after Luke Skywalker shoots him with a blaster round.
The scream that’s been used in more than 400 films is finding new life this year after California Institute of the Arts (CalArts) Professor Craig Smith discovered the Wilhelm’s original recording session while preserving a collection of 35-millimeter sound films he got from USC’s Cinematic Arts Library.
(17) SONIC SCREWDRIVER. There will be a new sonic screwdriver in the hands of the new Doctor Who. And it’s ready for its close-up.
Some of SpaceX’s Starlink satellites are leaking radio waves that could interfere with astronomy.
In the last few years, astronomers have warned about light pollution and other unintended consequences of the growing number of satellites. Since 2019, the company SpaceX in Hawthorne, California, has launched more than 4,300 Starlink broadband satellites into orbit, where they make up around half of all active satellites.
Federico Di Vruno at the Square Kilometre Array Observatory in Cheshire, UK, and his colleagues used the LOFAR radio telescope in the Netherlands to observe 68 Starlink satellites. They found that 47 of the satellites were emitting radio waves at frequencies very different from those used, and approved, for satellite communications with control stations on Earth.
The satellites’ emissions are not harming current radio astronomy observations, but such emissions might cause problems in future, as more and more satellites are launched. And although the emissions don’t violate any regulations, satellite operators and government authorities might consider regulating them, the authors say…
(19) FROM THE MAKERS OF MINIONS. The trailer for Migration, coming to theaters December 22.
The Mallard family is in a bit of rut. While dad Mack is content to keep his family safe paddling around their New England pond forever, mom Pam is eager to shake things up and show their kids—teen son Dax and duckling daughter Gwen—the whole wide world. After a migrating duck family alights on their pond with thrilling tales of far-flung places, Pam persuades Mack to embark on a family trip, via New York City, to tropical Jamaica. As the Mallards make their way South for the winter, their well-laid plans quickly go awry. The experience will inspire them to expand their horizons, open themselves up to new friends and accomplish more than they ever thought possible, while teaching them more about each other—and themselves—than they ever imagined.
[Thanks to Mike Kennedy, Andrew Porter, John King Tarpinian, Bill, Darrah Chavey, Steve French, Daniel Dern,Kimberly Unger, Chris Barkley, Michael Toman, Cat Eldridge, and SF Concatenation’s Jonathan Cowie for some of these stories. Title credit belongs to File 770 contributing editor of the day Jim Janney.]
34 thoughts on “Pixel Scroll 7/20/23 Crouching Pixel, Hidden Jetpack”
(8) Anyone here remember the BBC Oppenheimer biopic from 1980 or so? I watched it in high school (maybe my school got an educational copy that I watched in free periods?)
(11) Guy H. Lillian III – when I was 12-13 or so, in the mid 1970s, I wrote a letter to the editor of Superman or Action comics (don’t remember which). It didn’t get published, but I did get a postcard back from DC, written and signed by Mr. Lillian. I kept it pinned up on my bulletin board until I moved out for college. I wish I knew what happened to it . . . .
15) He’s extremely fortunate to still be alive.
(4) Uhm…
(15) Ugh. They call it “Death Valley” for a reason. Does no one remember what happened to Dave Legeno? 🙁
@ Andrew (not Werdna): I believe you may be looking for Day One, an excellent (and Emmy winning) 1989 BBC drama starring David Strathairn as J. Robert Oppenheimer and the late, great Brian Dennehy as Gen. Leslie Groves.
Pardon me, I MEANT that Day One was produced and broadcast on CBS, NOT the BBC…
CMB
(3) good for Coates. I’ve never read his equal for exact and powerful expression. His novel The Water Dancer (terrific) is genre, no?
(4) several of the fake electors claim they had no idea that their actions were illegal, after the secret meeting (phones confiscated to prevent recording) at which they were agreed, and some people refusing to sign because of doubting the legality. Amazing.
(8) I remember a PBS series with Sam Waterston as Oppenheimer. Somebody like David Suchet as Teller?
(3) Coates is truly remarkable. His ability to craft precise and impactful prose is unparalleled. I recently read his novel “The Water Dancer,” and I must say, it’s nothing short of terrific. Would you say it falls under a specific genre?
(4) It’s astonishing how some of the fake electors are claiming ignorance of the illegality of their actions. The fact that they held a secretive meeting, confiscating phones to prevent recording, and reached an agreement is quite concerning. And to make matters even more intriguing, some individuals refused to sign due to doubts about its legality. Absolutely amazing.
(8) I distinctly recall a PBS series featuring Sam Waterston portraying Oppenheimer. I wonder if someone like David Suchet would have been a suitable choice to play Teller in that series?
(15). Stupid, nothing but.
Andrew (not Werdna) on July 20, 2023 at 6:42 pm said:
(8) Anyone here remember the BBC Oppenheimer biopic from 1980 or so? I watched it in high school (maybe my school got an educational copy that I watched in free periods?)
Yes, I stayed up late to see it as a kid because I was obsessed with nuclear bombs at the time. Sam Waterston as the man himself and David Suchet as Edward Teller
1) These LLM-based chatbots can absolutely be set up to “cite their sources.” It would take serious effort, but again, there are log files somewhere of what the ‘bots were trained on, and since they are at base statistics engines, they can at least determine, statistically, where a response might come from. Even if it is more of a “60% likely to be made of 12% Chabon” than “this response is 72% derived from page 100 of Wonder Boys“, this is still a more honest answer than “too many sources to cite accurately.”
4) These fake electors knew exactly what they were doing. They just didn’t think the rules applied to them. Given that they committed real election fraud while trying to prevent imaginary election fraud (e.g. poor people an minorities being able to vote), this is an excellent example of Wilhoit’s Law: “Conservatism consists of exactly one proposition, to wit: There must be in-groups whom the law protects but does not bind, alongside out-groups whom the law binds but does not protect.”
Yes. It was the Waterston/Suchet version I watched. I wonder if I could find it on streaming
4) The law isn’t always right or moral. I can respect people who say, I’ll act according to my values regardless of what the law says. But ‘I didn’t know it was Illegal,’ when you clearly knew? Those people are ridiculous.
They need to read more sff to realize what they should do is have a daring prison escape and then hide out in the forest living in caves until the government is overthrown.
Diana Rigg was also in one of the wildest versions of Shakespeare’s “A Midsummer Night’s Dream,” in which she played Helena. Also in the stellar cast were Judi Dench, Helen Mirren, Ian Richardson , Ian Holm (as Puck), Clive Swift, and David Warner, among others.
Judi Dench as Tatiana, in green body paint, Diana Rigg and Helen Mirren in a catfight, and Clive Swift as the lion in the Pyramus and Thisbe “play within a play” are all wonderful.
Well worth watching!
“S’mores Than Human”
2) How many other Filers have already signed?
Andrew (not Werdna) asks Yes. It was the Waterston/Suchet version I watched. I wonder if I could find it on streaming
Now listening to Elizabeth Bear’s In the House of Aryaman, a Lonely Signal Burns
I’ve got the BBC Oppenheimer on DVD, errr, somewhere. (Which means it’s had at least one legitimate DVD release.) Good show, that, with lots of time spent on the political dimension, both domestic and international. I remember Clive James complaining that “the science has been left out”, but, c’mon, the BBC could hardly broadcast detailed instructions on how to make an atomic bomb, could they?
Carl Andor on July 21, 2023 at 7:13 am said:
Judi Dench as Tatiana, in green body paint,
She played She-Hulk!? 😀
(Sorry, I don’t normally pick on typos, but She-Hulk actually was played by Tatiana Maslany of Orphan Black fame, and I just saw an interview where Maslany jokingly introduced herself as Titania Miscellany, tying this joke back around.)
A belated thank you for the title credit.
Well, it seems I mixed up the 1989 CBS mini-series with the 1980 BBC production,which I remember being far superior. Here’s a article from Variety on it:
@ John W: “1) These LLM-based chatbots can absolutely be set up to “cite their sources.”
Do you have anything to back up that claim? It contradicts my understanding of how these programs work.
@Cliff–The bots’ makers know what they fed the bots. The bots are statistics engines. They can determine what those statistics engines drew from for responses, even if it’s not 100% certainty that “this part of this response is based on this specific page in X work.”
They can cite that.
And the companies can damn well pay the writers.
@Lis Carey: that doesn’t obviously follow. At the risk of fansplaining, consider a similar but simpler problem, autocomplete: someone has just typed a word, and you want to suggest a short list of probable next words. The usual way to do this is to construct a table. Each word has its own entry, whose value is a list of likely following words, something like this:
good -> luck morning night
to -> be the a
similar -> and to in
now -> that I is
may -> the I be
not -> to sure yet
Then when someone enters a word, you just look it up in the table and display the pre-constructed list. Where does the table come from? You write a program to scan a lot of texts and count how many times each word follows each other word, with the table as the program’s final output. Call it an autocomplete model. Here’s my point: nothing in the table contains any information about what texts were input or which part of which entry came from where. It would be possible to do that, but it would be considerably more difficult and the table would probably be exponentially larger; I can’t imagine anyone doing that unless it was a requirement.
@Jim Janney–The LLMs aren’t autocomplete. And what they do is not substantively like autocomplete. Autocomplete does not use writers’ creative content to–generate what looks like creative content. Too often, “creative content” that looks a lot like what the LLM was trained on.
Yes, the poor widdle programmers would have to do some more work. The companies are using their LLMs to build a business and make money. Which other commercial sources do they use things the sources charge for, without paying those commercial sources?
I’m betting they answer is “none.”
Pay the writers. They haven’t already done the work to make it easy to do so, because they were expecting to get away with stealing the material,
@Lis Carey: I also would like it to be that way, but so far no one has offered any evidence that it is that way, other than some handwaving about statistics engines, which as best I know do not work that way. Cliff made a respectful request for evidence, which has so far gone unanswered.
I know I’m being tedious and contentious, but the original claim was that existing chatbots can “absolutely” cite their sources.
When anyone says “this is how LLM’s work”, I take it with a grain of salt. Google and the other companies spend a pot of money developing them — why would they give specifics on what makes theirs better to random internet people? The details would be proprietary.
This is an excellent article discussing how ChatGTP works:
The secret sauce is usually the training corpus, training regimen, and the weights, not the architecture itself. Here’s the actual implementation of Llama: https://github.com/facebookresearch/llama/blob/main/llama/model.py but it’s not going to do anybody any good without the weights (which ended up getting leaked, but that’s another story) and good luck training it without facebook’s datasets and gpu clusters.
@Lis Carey:
They can determine what those statistics engines drew from for responses, even if it’s not 100% certainty that “this part of this response is based on this specific page in X work.”
I’ve been trying to think of ways that the models could incorporate attributions, and
the only thing I’ve come up with (have a parallel output layer that tries to predict input attribution for each token) would explode model size and probably wouldn’t work all that well either. Just the sheer magnitude of individual sources is overwhelming. (For example, there are hundreds of millions of active twitter accounts, and they’d all need their own individual attributions). It’d probably end up more like “there’s a 1-in-10000000 chance that any of these 10000000 sources were used”
I am in no way trying to justify these things, and would be perfectly happy if models that can’t avoid copyrighted content were prohibited from use.
Also
@Lis Carey
The LLMs aren’t autocomplete
They’re not the old fashioned style of autocomplete that just uses n-grams. Those were capable of “generat(ing) what looks like creative content” just with extremely low probability.
11: Diana Rigg played Vincent Price’s daughter in the horror movie Theatre of Blood. When the film was turned into a stage play, her own daughter Rachael Stirling played the same part.
Here’s a large language model that gives footnotes: Perplexity AI
(8) Anyone here remember the BBC Oppenheimer biopic from 1980 or so? I watched it in high school (maybe my school got an educational copy that I watched in free periods?)
(11) Guy H. Lillian III – when I was 12-13 or so, in the mid 1970s, I wrote a letter to the editor of Superman or Action comics (don’t remember which). It didn’t get published, but I did get a postcard back from DC, written and signed by Mr. Lillian. I kept it pinned up on my bulletin board until I moved out for college. I wish I knew what happened to it . . . .
15) He’s extremely fortunate to still be alive.
(4) Uhm…
(15) Ugh. They call it “Death Valley” for a reason. Does no one remember what happened to Dave Legeno? 🙁
@ Andrew (not Werdna): I believe you may be looking for Day One, an excellent (and Emmy winning) 1989 BBC drama starring David Strathairn as J. Robert Oppenheimer and the late, great Brian Dennehy as Gen. Leslie Groves.
https://en.wikipedia.org/wiki/Day_One_(1989_film)
CMB
Pardon me, I MEANT that Day One was produced and broadcast on CBS, NOT the BBC…
CMB
(3) good for Coates. I’ve never read his equal for exact and powerful expression. His novel The Water Dancer (terrific) is genre, no?
(4) several of the fake electors claim they had no idea that their actions were illegal, after the secret meeting (phones confiscated to prevent recording) at which they were agreed, and some people refusing to sign because of doubting the legality. Amazing.
(8) I remember a PBS series with Sam Waterston as Oppenheimer. Somebody like David Suchet as Teller?
(3) Coates is truly remarkable. His ability to craft precise and impactful prose is unparalleled. I recently read his novel “The Water Dancer,” and I must say, it’s nothing short of terrific. Would you say it falls under a specific genre?
(4) It’s astonishing how some of the fake electors are claiming ignorance of the illegality of their actions. The fact that they held a secretive meeting, confiscating phones to prevent recording, and reached an agreement is quite concerning. And to make matters even more intriguing, some individuals refused to sign due to doubts about its legality. Absolutely amazing.
(8) I distinctly recall a PBS series featuring Sam Waterston portraying Oppenheimer. I wonder if someone like David Suchet would have been a suitable choice to play Teller in that series?
(15). Stupid, nothing but.
Yes, I stayed up late to see it as a kid because I was obsessed with nuclear bombs at the time. Sam Waterston as the man himself and David Suchet as Edward Teller
https://www.imdb.com/title/tt0078037/
1) These LLM-based chatbots can absolutely be set up to “cite their sources.” It would take serious effort, but again, there are log files somewhere of what the ‘bots were trained on, and since they are at base statistics engines, they can at least determine, statistically, where a response might come from. Even if it is more of a “60% likely to be made of 12% Chabon” than “this response is 72% derived from page 100 of Wonder Boys“, this is still a more honest answer than “too many sources to cite accurately.”
4) These fake electors knew exactly what they were doing. They just didn’t think the rules applied to them. Given that they committed real election fraud while trying to prevent imaginary election fraud (e.g. poor people an minorities being able to vote), this is an excellent example of Wilhoit’s Law: “Conservatism consists of exactly one proposition, to wit: There must be in-groups whom the law protects but does not bind, alongside out-groups whom the law binds but does not protect.”
Yes. It was the Waterston/Suchet version I watched. I wonder if I could find it on streaming
4) The law isn’t always right or moral. I can respect people who say, I’ll act according to my values regardless of what the law says. But ‘I didn’t know it was Illegal,’ when you clearly knew? Those people are ridiculous.
They need to read more sff to realize what they should do is have a daring prison escape and then hide out in the forest living in caves until the government is overthrown.
Diana Rigg was also in one of the wildest versions of Shakespeare’s “A Midsummer Night’s Dream,” in which she played Helena. Also in the stellar cast were Judi Dench, Helen Mirren, Ian Richardson , Ian Holm (as Puck), Clive Swift, and David Warner, among others.
Judi Dench as Tatiana, in green body paint, Diana Rigg and Helen Mirren in a catfight, and Clive Swift as the lion in the Pyramus and Thisbe “play within a play” are all wonderful.
Well worth watching!
“S’mores Than Human”
2) How many other Filers have already signed?
Andrew (not Werdna) asks Yes. It was the Waterston/Suchet version I watched. I wonder if I could find it on streaming
It’s on YouTube. Here’s the first episode.
Thanks Cat!
Andrew (not Werdna) says to meThanks Cat!
You’re welcome!
Now listening to Elizabeth Bear’s In the House of Aryaman, a Lonely Signal Burns
I’ve got the BBC Oppenheimer on DVD, errr, somewhere. (Which means it’s had at least one legitimate DVD release.) Good show, that, with lots of time spent on the political dimension, both domestic and international. I remember Clive James complaining that “the science has been left out”, but, c’mon, the BBC could hardly broadcast detailed instructions on how to make an atomic bomb, could they?
Carl Andor on July 21, 2023 at 7:13 am said:
She played She-Hulk!? 😀
(Sorry, I don’t normally pick on typos, but She-Hulk actually was played by Tatiana Maslany of Orphan Black fame, and I just saw an interview where Maslany jokingly introduced herself as Titania Miscellany, tying this joke back around.)
A belated thank you for the title credit.
Well, it seems I mixed up the 1989 CBS mini-series with the 1980 BBC production,which I remember being far superior. Here’s a article from Variety on it:
https://variety.com/2023/tv/global/oppenheimer-bbc-christopher-nolan-sam-waterston-1235677307/#recipient_hashed=4a79c9f5fc8960a26049ea2bb94c43b1e8aa4177528a95b9eec575fa9bd93e52&recipient_salt=165fc6763660b488f46c783a2f165a4a9330120337543fb0c26b692098f24825
@ John W: “1) These LLM-based chatbots can absolutely be set up to “cite their sources.”
Do you have anything to back up that claim? It contradicts my understanding of how these programs work.
@Cliff–The bots’ makers know what they fed the bots. The bots are statistics engines. They can determine what those statistics engines drew from for responses, even if it’s not 100% certainty that “this part of this response is based on this specific page in X work.”
They can cite that.
And the companies can damn well pay the writers.
@Lis Carey: that doesn’t obviously follow. At the risk of fansplaining, consider a similar but simpler problem, autocomplete: someone has just typed a word, and you want to suggest a short list of probable next words. The usual way to do this is to construct a table. Each word has its own entry, whose value is a list of likely following words, something like this:
good -> luck morning night
to -> be the a
similar -> and to in
now -> that I is
may -> the I be
not -> to sure yet
Then when someone enters a word, you just look it up in the table and display the pre-constructed list. Where does the table come from? You write a program to scan a lot of texts and count how many times each word follows each other word, with the table as the program’s final output. Call it an autocomplete model. Here’s my point: nothing in the table contains any information about what texts were input or which part of which entry came from where. It would be possible to do that, but it would be considerably more difficult and the table would probably be exponentially larger; I can’t imagine anyone doing that unless it was a requirement.
@Jim Janney–The LLMs aren’t autocomplete. And what they do is not substantively like autocomplete. Autocomplete does not use writers’ creative content to–generate what looks like creative content. Too often, “creative content” that looks a lot like what the LLM was trained on.
Yes, the poor widdle programmers would have to do some more work. The companies are using their LLMs to build a business and make money. Which other commercial sources do they use things the sources charge for, without paying those commercial sources?
I’m betting they answer is “none.”
Pay the writers. They haven’t already done the work to make it easy to do so, because they were expecting to get away with stealing the material,
@Lis Carey: I also would like it to be that way, but so far no one has offered any evidence that it is that way, other than some handwaving about statistics engines, which as best I know do not work that way. Cliff made a respectful request for evidence, which has so far gone unanswered.
I know I’m being tedious and contentious, but the original claim was that existing chatbots can “absolutely” cite their sources.
When anyone says “this is how LLM’s work”, I take it with a grain of salt. Google and the other companies spend a pot of money developing them — why would they give specifics on what makes theirs better to random internet people? The details would be proprietary.
This is an excellent article discussing how ChatGTP works:
https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
Meta/Facebook open-sourced its own LLM a few days ago. The code and supporting documentation can be found here:
https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md
@Bill:
The secret sauce is usually the training corpus, training regimen, and the weights, not the architecture itself. Here’s the actual implementation of Llama: https://github.com/facebookresearch/llama/blob/main/llama/model.py but it’s not going to do anybody any good without the weights (which ended up getting leaked, but that’s another story) and good luck training it without facebook’s datasets and gpu clusters.
@Lis Carey:
I’ve been trying to think of ways that the models could incorporate attributions, and
the only thing I’ve come up with (have a parallel output layer that tries to predict input attribution for each token) would explode model size and probably wouldn’t work all that well either. Just the sheer magnitude of individual sources is overwhelming. (For example, there are hundreds of millions of active twitter accounts, and they’d all need their own individual attributions). It’d probably end up more like “there’s a 1-in-10000000 chance that any of these 10000000 sources were used”
I am in no way trying to justify these things, and would be perfectly happy if models that can’t avoid copyrighted content were prohibited from use.
Also
@Lis Carey
They’re not the old fashioned style of autocomplete that just uses n-grams. Those were capable of “generat(ing) what looks like creative content” just with extremely low probability.
11: Diana Rigg played Vincent Price’s daughter in the horror movie Theatre of Blood. When the film was turned into a stage play, her own daughter Rachael Stirling played the same part.
Here’s a large language model that gives footnotes: Perplexity AI